Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* SCTP kernel reference Implementation |
| 2 | * (C) Copyright IBM Corp. 2001, 2004 |
| 3 | * Copyright (c) 1999-2000 Cisco, Inc. |
| 4 | * Copyright (c) 1999-2001 Motorola, Inc. |
| 5 | * Copyright (c) 2001 Intel Corp. |
| 6 | * Copyright (c) 2001 Nokia, Inc. |
| 7 | * Copyright (c) 2001 La Monte H.P. Yarroll |
| 8 | * |
| 9 | * This abstraction carries sctp events to the ULP (sockets). |
| 10 | * |
| 11 | * The SCTP reference implementation is free software; |
| 12 | * you can redistribute it and/or modify it under the terms of |
| 13 | * the GNU General Public License as published by |
| 14 | * the Free Software Foundation; either version 2, or (at your option) |
| 15 | * any later version. |
| 16 | * |
| 17 | * The SCTP reference implementation is distributed in the hope that it |
| 18 | * will be useful, but WITHOUT ANY WARRANTY; without even the implied |
| 19 | * ************************ |
| 20 | * warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. |
| 21 | * See the GNU General Public License for more details. |
| 22 | * |
| 23 | * You should have received a copy of the GNU General Public License |
| 24 | * along with GNU CC; see the file COPYING. If not, write to |
| 25 | * the Free Software Foundation, 59 Temple Place - Suite 330, |
| 26 | * Boston, MA 02111-1307, USA. |
| 27 | * |
| 28 | * Please send any bug reports or fixes you make to the |
| 29 | * email address(es): |
| 30 | * lksctp developers <lksctp-developers@lists.sourceforge.net> |
| 31 | * |
| 32 | * Or submit a bug report through the following website: |
| 33 | * http://www.sf.net/projects/lksctp |
| 34 | * |
| 35 | * Written or modified by: |
| 36 | * Jon Grimm <jgrimm@us.ibm.com> |
| 37 | * La Monte H.P. Yarroll <piggy@acm.org> |
| 38 | * Sridhar Samudrala <sri@us.ibm.com> |
| 39 | * |
| 40 | * Any bugs reported given to us we will try to fix... any fixes shared will |
| 41 | * be incorporated into the next SCTP release. |
| 42 | */ |
| 43 | |
| 44 | #include <linux/types.h> |
| 45 | #include <linux/skbuff.h> |
| 46 | #include <net/sock.h> |
| 47 | #include <net/sctp/structs.h> |
| 48 | #include <net/sctp/sctp.h> |
| 49 | #include <net/sctp/sm.h> |
| 50 | |
| 51 | /* Forward declarations for internal helpers. */ |
| 52 | static struct sctp_ulpevent * sctp_ulpq_reasm(struct sctp_ulpq *ulpq, |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 53 | struct sctp_ulpevent *); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 54 | static struct sctp_ulpevent * sctp_ulpq_order(struct sctp_ulpq *, |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 55 | struct sctp_ulpevent *); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 56 | |
| 57 | /* 1st Level Abstractions */ |
| 58 | |
| 59 | /* Initialize a ULP queue from a block of memory. */ |
| 60 | struct sctp_ulpq *sctp_ulpq_init(struct sctp_ulpq *ulpq, |
| 61 | struct sctp_association *asoc) |
| 62 | { |
| 63 | memset(ulpq, 0, sizeof(struct sctp_ulpq)); |
| 64 | |
| 65 | ulpq->asoc = asoc; |
| 66 | skb_queue_head_init(&ulpq->reasm); |
| 67 | skb_queue_head_init(&ulpq->lobby); |
| 68 | ulpq->pd_mode = 0; |
| 69 | ulpq->malloced = 0; |
| 70 | |
| 71 | return ulpq; |
| 72 | } |
| 73 | |
| 74 | |
| 75 | /* Flush the reassembly and ordering queues. */ |
| 76 | static void sctp_ulpq_flush(struct sctp_ulpq *ulpq) |
| 77 | { |
| 78 | struct sk_buff *skb; |
| 79 | struct sctp_ulpevent *event; |
| 80 | |
| 81 | while ((skb = __skb_dequeue(&ulpq->lobby)) != NULL) { |
| 82 | event = sctp_skb2event(skb); |
| 83 | sctp_ulpevent_free(event); |
| 84 | } |
| 85 | |
| 86 | while ((skb = __skb_dequeue(&ulpq->reasm)) != NULL) { |
| 87 | event = sctp_skb2event(skb); |
| 88 | sctp_ulpevent_free(event); |
| 89 | } |
| 90 | |
| 91 | } |
| 92 | |
| 93 | /* Dispose of a ulpqueue. */ |
| 94 | void sctp_ulpq_free(struct sctp_ulpq *ulpq) |
| 95 | { |
| 96 | sctp_ulpq_flush(ulpq); |
| 97 | if (ulpq->malloced) |
| 98 | kfree(ulpq); |
| 99 | } |
| 100 | |
| 101 | /* Process an incoming DATA chunk. */ |
| 102 | int sctp_ulpq_tail_data(struct sctp_ulpq *ulpq, struct sctp_chunk *chunk, |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 103 | gfp_t gfp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 104 | { |
| 105 | struct sk_buff_head temp; |
| 106 | sctp_data_chunk_t *hdr; |
| 107 | struct sctp_ulpevent *event; |
| 108 | |
| 109 | hdr = (sctp_data_chunk_t *) chunk->chunk_hdr; |
| 110 | |
| 111 | /* Create an event from the incoming chunk. */ |
| 112 | event = sctp_ulpevent_make_rcvmsg(chunk->asoc, chunk, gfp); |
| 113 | if (!event) |
| 114 | return -ENOMEM; |
| 115 | |
| 116 | /* Do reassembly if needed. */ |
| 117 | event = sctp_ulpq_reasm(ulpq, event); |
| 118 | |
| 119 | /* Do ordering if needed. */ |
| 120 | if ((event) && (event->msg_flags & MSG_EOR)){ |
| 121 | /* Create a temporary list to collect chunks on. */ |
| 122 | skb_queue_head_init(&temp); |
| 123 | __skb_queue_tail(&temp, sctp_event2skb(event)); |
| 124 | |
| 125 | event = sctp_ulpq_order(ulpq, event); |
| 126 | } |
| 127 | |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 128 | /* Send event to the ULP. 'event' is the sctp_ulpevent for |
| 129 | * very first SKB on the 'temp' list. |
| 130 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 131 | if (event) |
| 132 | sctp_ulpq_tail_event(ulpq, event); |
| 133 | |
| 134 | return 0; |
| 135 | } |
| 136 | |
| 137 | /* Add a new event for propagation to the ULP. */ |
| 138 | /* Clear the partial delivery mode for this socket. Note: This |
| 139 | * assumes that no association is currently in partial delivery mode. |
| 140 | */ |
| 141 | int sctp_clear_pd(struct sock *sk) |
| 142 | { |
| 143 | struct sctp_sock *sp = sctp_sk(sk); |
| 144 | |
| 145 | sp->pd_mode = 0; |
| 146 | if (!skb_queue_empty(&sp->pd_lobby)) { |
| 147 | struct list_head *list; |
| 148 | sctp_skb_list_tail(&sp->pd_lobby, &sk->sk_receive_queue); |
| 149 | list = (struct list_head *)&sctp_sk(sk)->pd_lobby; |
| 150 | INIT_LIST_HEAD(list); |
| 151 | return 1; |
| 152 | } |
| 153 | return 0; |
| 154 | } |
| 155 | |
| 156 | /* Clear the pd_mode and restart any pending messages waiting for delivery. */ |
| 157 | static int sctp_ulpq_clear_pd(struct sctp_ulpq *ulpq) |
| 158 | { |
| 159 | ulpq->pd_mode = 0; |
| 160 | return sctp_clear_pd(ulpq->asoc->base.sk); |
| 161 | } |
| 162 | |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 163 | /* If the SKB of 'event' is on a list, it is the first such member |
| 164 | * of that list. |
| 165 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 166 | int sctp_ulpq_tail_event(struct sctp_ulpq *ulpq, struct sctp_ulpevent *event) |
| 167 | { |
| 168 | struct sock *sk = ulpq->asoc->base.sk; |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 169 | struct sk_buff_head *queue, *skb_list; |
| 170 | struct sk_buff *skb = sctp_event2skb(event); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 171 | int clear_pd = 0; |
| 172 | |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 173 | skb_list = (struct sk_buff_head *) skb->prev; |
| 174 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 175 | /* If the socket is just going to throw this away, do not |
| 176 | * even try to deliver it. |
| 177 | */ |
| 178 | if (sock_flag(sk, SOCK_DEAD) || (sk->sk_shutdown & RCV_SHUTDOWN)) |
| 179 | goto out_free; |
| 180 | |
| 181 | /* Check if the user wishes to receive this event. */ |
| 182 | if (!sctp_ulpevent_is_enabled(event, &sctp_sk(sk)->subscribe)) |
| 183 | goto out_free; |
| 184 | |
| 185 | /* If we are in partial delivery mode, post to the lobby until |
| 186 | * partial delivery is cleared, unless, of course _this_ is |
| 187 | * the association the cause of the partial delivery. |
| 188 | */ |
| 189 | |
| 190 | if (!sctp_sk(sk)->pd_mode) { |
| 191 | queue = &sk->sk_receive_queue; |
| 192 | } else if (ulpq->pd_mode) { |
| 193 | if (event->msg_flags & MSG_NOTIFICATION) |
| 194 | queue = &sctp_sk(sk)->pd_lobby; |
| 195 | else { |
| 196 | clear_pd = event->msg_flags & MSG_EOR; |
| 197 | queue = &sk->sk_receive_queue; |
| 198 | } |
| 199 | } else |
| 200 | queue = &sctp_sk(sk)->pd_lobby; |
| 201 | |
| 202 | |
| 203 | /* If we are harvesting multiple skbs they will be |
| 204 | * collected on a list. |
| 205 | */ |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 206 | if (skb_list) |
| 207 | sctp_skb_list_tail(skb_list, queue); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 208 | else |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 209 | __skb_queue_tail(queue, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 210 | |
| 211 | /* Did we just complete partial delivery and need to get |
| 212 | * rolling again? Move pending data to the receive |
| 213 | * queue. |
| 214 | */ |
| 215 | if (clear_pd) |
| 216 | sctp_ulpq_clear_pd(ulpq); |
| 217 | |
| 218 | if (queue == &sk->sk_receive_queue) |
| 219 | sk->sk_data_ready(sk, 0); |
| 220 | return 1; |
| 221 | |
| 222 | out_free: |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 223 | if (skb_list) |
| 224 | sctp_queue_purge_ulpevents(skb_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 225 | else |
| 226 | sctp_ulpevent_free(event); |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 227 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 228 | return 0; |
| 229 | } |
| 230 | |
| 231 | /* 2nd Level Abstractions */ |
| 232 | |
| 233 | /* Helper function to store chunks that need to be reassembled. */ |
| 234 | static inline void sctp_ulpq_store_reasm(struct sctp_ulpq *ulpq, |
| 235 | struct sctp_ulpevent *event) |
| 236 | { |
| 237 | struct sk_buff *pos; |
| 238 | struct sctp_ulpevent *cevent; |
| 239 | __u32 tsn, ctsn; |
| 240 | |
| 241 | tsn = event->tsn; |
| 242 | |
| 243 | /* See if it belongs at the end. */ |
| 244 | pos = skb_peek_tail(&ulpq->reasm); |
| 245 | if (!pos) { |
| 246 | __skb_queue_tail(&ulpq->reasm, sctp_event2skb(event)); |
| 247 | return; |
| 248 | } |
| 249 | |
| 250 | /* Short circuit just dropping it at the end. */ |
| 251 | cevent = sctp_skb2event(pos); |
| 252 | ctsn = cevent->tsn; |
| 253 | if (TSN_lt(ctsn, tsn)) { |
| 254 | __skb_queue_tail(&ulpq->reasm, sctp_event2skb(event)); |
| 255 | return; |
| 256 | } |
| 257 | |
| 258 | /* Find the right place in this list. We store them by TSN. */ |
| 259 | skb_queue_walk(&ulpq->reasm, pos) { |
| 260 | cevent = sctp_skb2event(pos); |
| 261 | ctsn = cevent->tsn; |
| 262 | |
| 263 | if (TSN_lt(tsn, ctsn)) |
| 264 | break; |
| 265 | } |
| 266 | |
| 267 | /* Insert before pos. */ |
| 268 | __skb_insert(sctp_event2skb(event), pos->prev, pos, &ulpq->reasm); |
| 269 | |
| 270 | } |
| 271 | |
| 272 | /* Helper function to return an event corresponding to the reassembled |
| 273 | * datagram. |
| 274 | * This routine creates a re-assembled skb given the first and last skb's |
| 275 | * as stored in the reassembly queue. The skb's may be non-linear if the sctp |
| 276 | * payload was fragmented on the way and ip had to reassemble them. |
| 277 | * We add the rest of skb's to the first skb's fraglist. |
| 278 | */ |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 279 | static struct sctp_ulpevent *sctp_make_reassembled_event(struct sk_buff_head *queue, struct sk_buff *f_frag, struct sk_buff *l_frag) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 280 | { |
| 281 | struct sk_buff *pos; |
Vladislav Yasevich | 672e7cc | 2006-05-05 17:03:49 -0700 | [diff] [blame] | 282 | struct sk_buff *new = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 283 | struct sctp_ulpevent *event; |
| 284 | struct sk_buff *pnext, *last; |
| 285 | struct sk_buff *list = skb_shinfo(f_frag)->frag_list; |
| 286 | |
| 287 | /* Store the pointer to the 2nd skb */ |
| 288 | if (f_frag == l_frag) |
| 289 | pos = NULL; |
| 290 | else |
| 291 | pos = f_frag->next; |
| 292 | |
| 293 | /* Get the last skb in the f_frag's frag_list if present. */ |
| 294 | for (last = list; list; last = list, list = list->next); |
| 295 | |
| 296 | /* Add the list of remaining fragments to the first fragments |
| 297 | * frag_list. |
| 298 | */ |
| 299 | if (last) |
| 300 | last->next = pos; |
Vladislav Yasevich | 672e7cc | 2006-05-05 17:03:49 -0700 | [diff] [blame] | 301 | else { |
| 302 | if (skb_cloned(f_frag)) { |
| 303 | /* This is a cloned skb, we can't just modify |
| 304 | * the frag_list. We need a new skb to do that. |
| 305 | * Instead of calling skb_unshare(), we'll do it |
| 306 | * ourselves since we need to delay the free. |
| 307 | */ |
| 308 | new = skb_copy(f_frag, GFP_ATOMIC); |
| 309 | if (!new) |
| 310 | return NULL; /* try again later */ |
| 311 | |
| 312 | new->sk = f_frag->sk; |
| 313 | |
| 314 | skb_shinfo(new)->frag_list = pos; |
| 315 | } else |
| 316 | skb_shinfo(f_frag)->frag_list = pos; |
| 317 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 318 | |
| 319 | /* Remove the first fragment from the reassembly queue. */ |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 320 | __skb_unlink(f_frag, queue); |
Vladislav Yasevich | 672e7cc | 2006-05-05 17:03:49 -0700 | [diff] [blame] | 321 | |
| 322 | /* if we did unshare, then free the old skb and re-assign */ |
| 323 | if (new) { |
| 324 | kfree_skb(f_frag); |
| 325 | f_frag = new; |
| 326 | } |
| 327 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 328 | while (pos) { |
| 329 | |
| 330 | pnext = pos->next; |
| 331 | |
| 332 | /* Update the len and data_len fields of the first fragment. */ |
| 333 | f_frag->len += pos->len; |
| 334 | f_frag->data_len += pos->len; |
| 335 | |
| 336 | /* Remove the fragment from the reassembly queue. */ |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 337 | __skb_unlink(pos, queue); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 338 | |
| 339 | /* Break if we have reached the last fragment. */ |
| 340 | if (pos == l_frag) |
| 341 | break; |
| 342 | pos->next = pnext; |
| 343 | pos = pnext; |
| 344 | }; |
| 345 | |
| 346 | event = sctp_skb2event(f_frag); |
| 347 | SCTP_INC_STATS(SCTP_MIB_REASMUSRMSGS); |
| 348 | |
| 349 | return event; |
| 350 | } |
| 351 | |
| 352 | |
| 353 | /* Helper function to check if an incoming chunk has filled up the last |
| 354 | * missing fragment in a SCTP datagram and return the corresponding event. |
| 355 | */ |
| 356 | static inline struct sctp_ulpevent *sctp_ulpq_retrieve_reassembled(struct sctp_ulpq *ulpq) |
| 357 | { |
| 358 | struct sk_buff *pos; |
| 359 | struct sctp_ulpevent *cevent; |
| 360 | struct sk_buff *first_frag = NULL; |
| 361 | __u32 ctsn, next_tsn; |
| 362 | struct sctp_ulpevent *retval = NULL; |
| 363 | |
| 364 | /* Initialized to 0 just to avoid compiler warning message. Will |
| 365 | * never be used with this value. It is referenced only after it |
| 366 | * is set when we find the first fragment of a message. |
| 367 | */ |
| 368 | next_tsn = 0; |
| 369 | |
| 370 | /* The chunks are held in the reasm queue sorted by TSN. |
| 371 | * Walk through the queue sequentially and look for a sequence of |
| 372 | * fragmented chunks that complete a datagram. |
| 373 | * 'first_frag' and next_tsn are reset when we find a chunk which |
| 374 | * is the first fragment of a datagram. Once these 2 fields are set |
| 375 | * we expect to find the remaining middle fragments and the last |
| 376 | * fragment in order. If not, first_frag is reset to NULL and we |
| 377 | * start the next pass when we find another first fragment. |
| 378 | */ |
| 379 | skb_queue_walk(&ulpq->reasm, pos) { |
| 380 | cevent = sctp_skb2event(pos); |
| 381 | ctsn = cevent->tsn; |
| 382 | |
| 383 | switch (cevent->msg_flags & SCTP_DATA_FRAG_MASK) { |
| 384 | case SCTP_DATA_FIRST_FRAG: |
| 385 | first_frag = pos; |
| 386 | next_tsn = ctsn + 1; |
| 387 | break; |
| 388 | |
| 389 | case SCTP_DATA_MIDDLE_FRAG: |
| 390 | if ((first_frag) && (ctsn == next_tsn)) |
| 391 | next_tsn++; |
| 392 | else |
| 393 | first_frag = NULL; |
| 394 | break; |
| 395 | |
| 396 | case SCTP_DATA_LAST_FRAG: |
| 397 | if (first_frag && (ctsn == next_tsn)) |
| 398 | goto found; |
| 399 | else |
| 400 | first_frag = NULL; |
| 401 | break; |
| 402 | }; |
| 403 | |
| 404 | } |
| 405 | done: |
| 406 | return retval; |
| 407 | found: |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 408 | retval = sctp_make_reassembled_event(&ulpq->reasm, first_frag, pos); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 409 | if (retval) |
| 410 | retval->msg_flags |= MSG_EOR; |
| 411 | goto done; |
| 412 | } |
| 413 | |
| 414 | /* Retrieve the next set of fragments of a partial message. */ |
| 415 | static inline struct sctp_ulpevent *sctp_ulpq_retrieve_partial(struct sctp_ulpq *ulpq) |
| 416 | { |
| 417 | struct sk_buff *pos, *last_frag, *first_frag; |
| 418 | struct sctp_ulpevent *cevent; |
| 419 | __u32 ctsn, next_tsn; |
| 420 | int is_last; |
| 421 | struct sctp_ulpevent *retval; |
| 422 | |
| 423 | /* The chunks are held in the reasm queue sorted by TSN. |
| 424 | * Walk through the queue sequentially and look for the first |
| 425 | * sequence of fragmented chunks. |
| 426 | */ |
| 427 | |
| 428 | if (skb_queue_empty(&ulpq->reasm)) |
| 429 | return NULL; |
| 430 | |
| 431 | last_frag = first_frag = NULL; |
| 432 | retval = NULL; |
| 433 | next_tsn = 0; |
| 434 | is_last = 0; |
| 435 | |
| 436 | skb_queue_walk(&ulpq->reasm, pos) { |
| 437 | cevent = sctp_skb2event(pos); |
| 438 | ctsn = cevent->tsn; |
| 439 | |
| 440 | switch (cevent->msg_flags & SCTP_DATA_FRAG_MASK) { |
| 441 | case SCTP_DATA_MIDDLE_FRAG: |
| 442 | if (!first_frag) { |
| 443 | first_frag = pos; |
| 444 | next_tsn = ctsn + 1; |
| 445 | last_frag = pos; |
| 446 | } else if (next_tsn == ctsn) |
| 447 | next_tsn++; |
| 448 | else |
| 449 | goto done; |
| 450 | break; |
| 451 | case SCTP_DATA_LAST_FRAG: |
| 452 | if (!first_frag) |
| 453 | first_frag = pos; |
| 454 | else if (ctsn != next_tsn) |
| 455 | goto done; |
| 456 | last_frag = pos; |
| 457 | is_last = 1; |
| 458 | goto done; |
| 459 | default: |
| 460 | return NULL; |
| 461 | }; |
| 462 | } |
| 463 | |
| 464 | /* We have the reassembled event. There is no need to look |
| 465 | * further. |
| 466 | */ |
| 467 | done: |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 468 | retval = sctp_make_reassembled_event(&ulpq->reasm, first_frag, last_frag); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 469 | if (retval && is_last) |
| 470 | retval->msg_flags |= MSG_EOR; |
| 471 | |
| 472 | return retval; |
| 473 | } |
| 474 | |
| 475 | |
| 476 | /* Helper function to reassemble chunks. Hold chunks on the reasm queue that |
| 477 | * need reassembling. |
| 478 | */ |
| 479 | static struct sctp_ulpevent *sctp_ulpq_reasm(struct sctp_ulpq *ulpq, |
| 480 | struct sctp_ulpevent *event) |
| 481 | { |
| 482 | struct sctp_ulpevent *retval = NULL; |
| 483 | |
| 484 | /* Check if this is part of a fragmented message. */ |
| 485 | if (SCTP_DATA_NOT_FRAG == (event->msg_flags & SCTP_DATA_FRAG_MASK)) { |
| 486 | event->msg_flags |= MSG_EOR; |
| 487 | return event; |
| 488 | } |
| 489 | |
| 490 | sctp_ulpq_store_reasm(ulpq, event); |
| 491 | if (!ulpq->pd_mode) |
| 492 | retval = sctp_ulpq_retrieve_reassembled(ulpq); |
| 493 | else { |
| 494 | __u32 ctsn, ctsnap; |
| 495 | |
| 496 | /* Do not even bother unless this is the next tsn to |
| 497 | * be delivered. |
| 498 | */ |
| 499 | ctsn = event->tsn; |
| 500 | ctsnap = sctp_tsnmap_get_ctsn(&ulpq->asoc->peer.tsn_map); |
| 501 | if (TSN_lte(ctsn, ctsnap)) |
| 502 | retval = sctp_ulpq_retrieve_partial(ulpq); |
| 503 | } |
| 504 | |
| 505 | return retval; |
| 506 | } |
| 507 | |
| 508 | /* Retrieve the first part (sequential fragments) for partial delivery. */ |
| 509 | static inline struct sctp_ulpevent *sctp_ulpq_retrieve_first(struct sctp_ulpq *ulpq) |
| 510 | { |
| 511 | struct sk_buff *pos, *last_frag, *first_frag; |
| 512 | struct sctp_ulpevent *cevent; |
| 513 | __u32 ctsn, next_tsn; |
| 514 | struct sctp_ulpevent *retval; |
| 515 | |
| 516 | /* The chunks are held in the reasm queue sorted by TSN. |
| 517 | * Walk through the queue sequentially and look for a sequence of |
| 518 | * fragmented chunks that start a datagram. |
| 519 | */ |
| 520 | |
| 521 | if (skb_queue_empty(&ulpq->reasm)) |
| 522 | return NULL; |
| 523 | |
| 524 | last_frag = first_frag = NULL; |
| 525 | retval = NULL; |
| 526 | next_tsn = 0; |
| 527 | |
| 528 | skb_queue_walk(&ulpq->reasm, pos) { |
| 529 | cevent = sctp_skb2event(pos); |
| 530 | ctsn = cevent->tsn; |
| 531 | |
| 532 | switch (cevent->msg_flags & SCTP_DATA_FRAG_MASK) { |
| 533 | case SCTP_DATA_FIRST_FRAG: |
| 534 | if (!first_frag) { |
| 535 | first_frag = pos; |
| 536 | next_tsn = ctsn + 1; |
| 537 | last_frag = pos; |
| 538 | } else |
| 539 | goto done; |
| 540 | break; |
| 541 | |
| 542 | case SCTP_DATA_MIDDLE_FRAG: |
| 543 | if (!first_frag) |
| 544 | return NULL; |
| 545 | if (ctsn == next_tsn) { |
| 546 | next_tsn++; |
| 547 | last_frag = pos; |
| 548 | } else |
| 549 | goto done; |
| 550 | break; |
| 551 | default: |
| 552 | return NULL; |
| 553 | }; |
| 554 | } |
| 555 | |
| 556 | /* We have the reassembled event. There is no need to look |
| 557 | * further. |
| 558 | */ |
| 559 | done: |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 560 | retval = sctp_make_reassembled_event(&ulpq->reasm, first_frag, last_frag); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 561 | return retval; |
| 562 | } |
| 563 | |
| 564 | /* Helper function to gather skbs that have possibly become |
| 565 | * ordered by an an incoming chunk. |
| 566 | */ |
| 567 | static inline void sctp_ulpq_retrieve_ordered(struct sctp_ulpq *ulpq, |
| 568 | struct sctp_ulpevent *event) |
| 569 | { |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 570 | struct sk_buff_head *event_list; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 571 | struct sk_buff *pos, *tmp; |
| 572 | struct sctp_ulpevent *cevent; |
| 573 | struct sctp_stream *in; |
| 574 | __u16 sid, csid; |
| 575 | __u16 ssn, cssn; |
| 576 | |
| 577 | sid = event->stream; |
| 578 | ssn = event->ssn; |
| 579 | in = &ulpq->asoc->ssnmap->in; |
| 580 | |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 581 | event_list = (struct sk_buff_head *) sctp_event2skb(event)->prev; |
| 582 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 583 | /* We are holding the chunks by stream, by SSN. */ |
| 584 | sctp_skb_for_each(pos, &ulpq->lobby, tmp) { |
| 585 | cevent = (struct sctp_ulpevent *) pos->cb; |
| 586 | csid = cevent->stream; |
| 587 | cssn = cevent->ssn; |
| 588 | |
| 589 | /* Have we gone too far? */ |
| 590 | if (csid > sid) |
| 591 | break; |
| 592 | |
| 593 | /* Have we not gone far enough? */ |
| 594 | if (csid < sid) |
| 595 | continue; |
| 596 | |
| 597 | if (cssn != sctp_ssn_peek(in, sid)) |
| 598 | break; |
| 599 | |
| 600 | /* Found it, so mark in the ssnmap. */ |
| 601 | sctp_ssn_next(in, sid); |
| 602 | |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 603 | __skb_unlink(pos, &ulpq->lobby); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 604 | |
| 605 | /* Attach all gathered skbs to the event. */ |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 606 | __skb_queue_tail(event_list, pos); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 607 | } |
| 608 | } |
| 609 | |
| 610 | /* Helper function to store chunks needing ordering. */ |
| 611 | static inline void sctp_ulpq_store_ordered(struct sctp_ulpq *ulpq, |
| 612 | struct sctp_ulpevent *event) |
| 613 | { |
| 614 | struct sk_buff *pos; |
| 615 | struct sctp_ulpevent *cevent; |
| 616 | __u16 sid, csid; |
| 617 | __u16 ssn, cssn; |
| 618 | |
| 619 | pos = skb_peek_tail(&ulpq->lobby); |
| 620 | if (!pos) { |
| 621 | __skb_queue_tail(&ulpq->lobby, sctp_event2skb(event)); |
| 622 | return; |
| 623 | } |
| 624 | |
| 625 | sid = event->stream; |
| 626 | ssn = event->ssn; |
| 627 | |
| 628 | cevent = (struct sctp_ulpevent *) pos->cb; |
| 629 | csid = cevent->stream; |
| 630 | cssn = cevent->ssn; |
| 631 | if (sid > csid) { |
| 632 | __skb_queue_tail(&ulpq->lobby, sctp_event2skb(event)); |
| 633 | return; |
| 634 | } |
| 635 | |
| 636 | if ((sid == csid) && SSN_lt(cssn, ssn)) { |
| 637 | __skb_queue_tail(&ulpq->lobby, sctp_event2skb(event)); |
| 638 | return; |
| 639 | } |
| 640 | |
| 641 | /* Find the right place in this list. We store them by |
| 642 | * stream ID and then by SSN. |
| 643 | */ |
| 644 | skb_queue_walk(&ulpq->lobby, pos) { |
| 645 | cevent = (struct sctp_ulpevent *) pos->cb; |
| 646 | csid = cevent->stream; |
| 647 | cssn = cevent->ssn; |
| 648 | |
| 649 | if (csid > sid) |
| 650 | break; |
| 651 | if (csid == sid && SSN_lt(ssn, cssn)) |
| 652 | break; |
| 653 | } |
| 654 | |
| 655 | |
| 656 | /* Insert before pos. */ |
| 657 | __skb_insert(sctp_event2skb(event), pos->prev, pos, &ulpq->lobby); |
| 658 | |
| 659 | } |
| 660 | |
| 661 | static struct sctp_ulpevent *sctp_ulpq_order(struct sctp_ulpq *ulpq, |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 662 | struct sctp_ulpevent *event) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 663 | { |
| 664 | __u16 sid, ssn; |
| 665 | struct sctp_stream *in; |
| 666 | |
| 667 | /* Check if this message needs ordering. */ |
| 668 | if (SCTP_DATA_UNORDERED & event->msg_flags) |
| 669 | return event; |
| 670 | |
| 671 | /* Note: The stream ID must be verified before this routine. */ |
| 672 | sid = event->stream; |
| 673 | ssn = event->ssn; |
| 674 | in = &ulpq->asoc->ssnmap->in; |
| 675 | |
| 676 | /* Is this the expected SSN for this stream ID? */ |
| 677 | if (ssn != sctp_ssn_peek(in, sid)) { |
| 678 | /* We've received something out of order, so find where it |
| 679 | * needs to be placed. We order by stream and then by SSN. |
| 680 | */ |
| 681 | sctp_ulpq_store_ordered(ulpq, event); |
| 682 | return NULL; |
| 683 | } |
| 684 | |
| 685 | /* Mark that the next chunk has been found. */ |
| 686 | sctp_ssn_next(in, sid); |
| 687 | |
| 688 | /* Go find any other chunks that were waiting for |
| 689 | * ordering. |
| 690 | */ |
| 691 | sctp_ulpq_retrieve_ordered(ulpq, event); |
| 692 | |
| 693 | return event; |
| 694 | } |
| 695 | |
| 696 | /* Helper function to gather skbs that have possibly become |
| 697 | * ordered by forward tsn skipping their dependencies. |
| 698 | */ |
| 699 | static inline void sctp_ulpq_reap_ordered(struct sctp_ulpq *ulpq) |
| 700 | { |
| 701 | struct sk_buff *pos, *tmp; |
| 702 | struct sctp_ulpevent *cevent; |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 703 | struct sctp_ulpevent *event; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 704 | struct sctp_stream *in; |
| 705 | struct sk_buff_head temp; |
| 706 | __u16 csid, cssn; |
| 707 | |
| 708 | in = &ulpq->asoc->ssnmap->in; |
| 709 | |
| 710 | /* We are holding the chunks by stream, by SSN. */ |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 711 | skb_queue_head_init(&temp); |
| 712 | event = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 713 | sctp_skb_for_each(pos, &ulpq->lobby, tmp) { |
| 714 | cevent = (struct sctp_ulpevent *) pos->cb; |
| 715 | csid = cevent->stream; |
| 716 | cssn = cevent->ssn; |
| 717 | |
| 718 | if (cssn != sctp_ssn_peek(in, csid)) |
| 719 | break; |
| 720 | |
| 721 | /* Found it, so mark in the ssnmap. */ |
| 722 | sctp_ssn_next(in, csid); |
| 723 | |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 724 | __skb_unlink(pos, &ulpq->lobby); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 725 | if (!event) { |
| 726 | /* Create a temporary list to collect chunks on. */ |
| 727 | event = sctp_skb2event(pos); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 728 | __skb_queue_tail(&temp, sctp_event2skb(event)); |
| 729 | } else { |
| 730 | /* Attach all gathered skbs to the event. */ |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 731 | __skb_queue_tail(&temp, pos); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 732 | } |
| 733 | } |
| 734 | |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 735 | /* Send event to the ULP. 'event' is the sctp_ulpevent for |
| 736 | * very first SKB on the 'temp' list. |
| 737 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 738 | if (event) |
| 739 | sctp_ulpq_tail_event(ulpq, event); |
| 740 | } |
| 741 | |
| 742 | /* Skip over an SSN. */ |
| 743 | void sctp_ulpq_skip(struct sctp_ulpq *ulpq, __u16 sid, __u16 ssn) |
| 744 | { |
| 745 | struct sctp_stream *in; |
| 746 | |
| 747 | /* Note: The stream ID must be verified before this routine. */ |
| 748 | in = &ulpq->asoc->ssnmap->in; |
| 749 | |
| 750 | /* Is this an old SSN? If so ignore. */ |
| 751 | if (SSN_lt(ssn, sctp_ssn_peek(in, sid))) |
| 752 | return; |
| 753 | |
| 754 | /* Mark that we are no longer expecting this SSN or lower. */ |
| 755 | sctp_ssn_skip(in, sid, ssn); |
| 756 | |
| 757 | /* Go find any other chunks that were waiting for |
| 758 | * ordering and deliver them if needed. |
| 759 | */ |
| 760 | sctp_ulpq_reap_ordered(ulpq); |
| 761 | return; |
| 762 | } |
| 763 | |
| 764 | /* Renege 'needed' bytes from the ordering queue. */ |
| 765 | static __u16 sctp_ulpq_renege_order(struct sctp_ulpq *ulpq, __u16 needed) |
| 766 | { |
| 767 | __u16 freed = 0; |
| 768 | __u32 tsn; |
| 769 | struct sk_buff *skb; |
| 770 | struct sctp_ulpevent *event; |
| 771 | struct sctp_tsnmap *tsnmap; |
| 772 | |
| 773 | tsnmap = &ulpq->asoc->peer.tsn_map; |
| 774 | |
| 775 | while ((skb = __skb_dequeue_tail(&ulpq->lobby)) != NULL) { |
| 776 | freed += skb_headlen(skb); |
| 777 | event = sctp_skb2event(skb); |
| 778 | tsn = event->tsn; |
| 779 | |
| 780 | sctp_ulpevent_free(event); |
| 781 | sctp_tsnmap_renege(tsnmap, tsn); |
| 782 | if (freed >= needed) |
| 783 | return freed; |
| 784 | } |
| 785 | |
| 786 | return freed; |
| 787 | } |
| 788 | |
| 789 | /* Renege 'needed' bytes from the reassembly queue. */ |
| 790 | static __u16 sctp_ulpq_renege_frags(struct sctp_ulpq *ulpq, __u16 needed) |
| 791 | { |
| 792 | __u16 freed = 0; |
| 793 | __u32 tsn; |
| 794 | struct sk_buff *skb; |
| 795 | struct sctp_ulpevent *event; |
| 796 | struct sctp_tsnmap *tsnmap; |
| 797 | |
| 798 | tsnmap = &ulpq->asoc->peer.tsn_map; |
| 799 | |
| 800 | /* Walk backwards through the list, reneges the newest tsns. */ |
| 801 | while ((skb = __skb_dequeue_tail(&ulpq->reasm)) != NULL) { |
| 802 | freed += skb_headlen(skb); |
| 803 | event = sctp_skb2event(skb); |
| 804 | tsn = event->tsn; |
| 805 | |
| 806 | sctp_ulpevent_free(event); |
| 807 | sctp_tsnmap_renege(tsnmap, tsn); |
| 808 | if (freed >= needed) |
| 809 | return freed; |
| 810 | } |
| 811 | |
| 812 | return freed; |
| 813 | } |
| 814 | |
| 815 | /* Partial deliver the first message as there is pressure on rwnd. */ |
| 816 | void sctp_ulpq_partial_delivery(struct sctp_ulpq *ulpq, |
Alexey Dobriyan | 3182cd8 | 2005-07-11 20:57:47 -0700 | [diff] [blame] | 817 | struct sctp_chunk *chunk, |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 818 | gfp_t gfp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 819 | { |
| 820 | struct sctp_ulpevent *event; |
| 821 | struct sctp_association *asoc; |
| 822 | |
| 823 | asoc = ulpq->asoc; |
| 824 | |
| 825 | /* Are we already in partial delivery mode? */ |
| 826 | if (!sctp_sk(asoc->base.sk)->pd_mode) { |
| 827 | |
| 828 | /* Is partial delivery possible? */ |
| 829 | event = sctp_ulpq_retrieve_first(ulpq); |
| 830 | /* Send event to the ULP. */ |
| 831 | if (event) { |
| 832 | sctp_ulpq_tail_event(ulpq, event); |
| 833 | sctp_sk(asoc->base.sk)->pd_mode = 1; |
| 834 | ulpq->pd_mode = 1; |
| 835 | return; |
| 836 | } |
| 837 | } |
| 838 | } |
| 839 | |
| 840 | /* Renege some packets to make room for an incoming chunk. */ |
| 841 | void sctp_ulpq_renege(struct sctp_ulpq *ulpq, struct sctp_chunk *chunk, |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 842 | gfp_t gfp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 843 | { |
| 844 | struct sctp_association *asoc; |
| 845 | __u16 needed, freed; |
| 846 | |
| 847 | asoc = ulpq->asoc; |
| 848 | |
| 849 | if (chunk) { |
| 850 | needed = ntohs(chunk->chunk_hdr->length); |
| 851 | needed -= sizeof(sctp_data_chunk_t); |
| 852 | } else |
| 853 | needed = SCTP_DEFAULT_MAXWINDOW; |
| 854 | |
| 855 | freed = 0; |
| 856 | |
| 857 | if (skb_queue_empty(&asoc->base.sk->sk_receive_queue)) { |
| 858 | freed = sctp_ulpq_renege_order(ulpq, needed); |
| 859 | if (freed < needed) { |
| 860 | freed += sctp_ulpq_renege_frags(ulpq, needed - freed); |
| 861 | } |
| 862 | } |
| 863 | /* If able to free enough room, accept this chunk. */ |
| 864 | if (chunk && (freed >= needed)) { |
| 865 | __u32 tsn; |
| 866 | tsn = ntohl(chunk->subh.data_hdr->tsn); |
| 867 | sctp_tsnmap_mark(&asoc->peer.tsn_map, tsn); |
| 868 | sctp_ulpq_tail_data(ulpq, chunk, gfp); |
| 869 | |
| 870 | sctp_ulpq_partial_delivery(ulpq, chunk, gfp); |
| 871 | } |
| 872 | |
| 873 | return; |
| 874 | } |
| 875 | |
| 876 | |
| 877 | |
| 878 | /* Notify the application if an association is aborted and in |
| 879 | * partial delivery mode. Send up any pending received messages. |
| 880 | */ |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 881 | void sctp_ulpq_abort_pd(struct sctp_ulpq *ulpq, gfp_t gfp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 882 | { |
| 883 | struct sctp_ulpevent *ev = NULL; |
| 884 | struct sock *sk; |
| 885 | |
| 886 | if (!ulpq->pd_mode) |
| 887 | return; |
| 888 | |
| 889 | sk = ulpq->asoc->base.sk; |
| 890 | if (sctp_ulpevent_type_enabled(SCTP_PARTIAL_DELIVERY_EVENT, |
| 891 | &sctp_sk(sk)->subscribe)) |
| 892 | ev = sctp_ulpevent_make_pdapi(ulpq->asoc, |
| 893 | SCTP_PARTIAL_DELIVERY_ABORTED, |
| 894 | gfp); |
| 895 | if (ev) |
| 896 | __skb_queue_tail(&sk->sk_receive_queue, sctp_event2skb(ev)); |
| 897 | |
| 898 | /* If there is data waiting, send it up the socket now. */ |
| 899 | if (sctp_ulpq_clear_pd(ulpq) || ev) |
| 900 | sk->sk_data_ready(sk, 0); |
| 901 | } |